From 90fa8de96cdaf75e7fd352162fa3b0d7af724db5 Mon Sep 17 00:00:00 2001 From: Mandar Nilange Date: Fri, 1 May 2026 22:35:13 +0530 Subject: [PATCH 1/2] docs: bullet-ize README hero; align platform README with new framing Main README: - Replace the dense paragraph after the hero with five scannable bullets organized around the verbs that carry the pitch: Compose, Gate, Plug in, Run anywhere, Scale. - Tighten the trailing "domain-agnostic" line. - Hero one-liner and "For platform teams" K8s callout unchanged. Platform README: - Tie the lead to the new framework story ("Run agentic workflows at scale") and reference core as "the open framework for agentic workflows". - Make the BYO framing explicit: bring your own LLMs, scripts, agents, and infra; this package handles the orchestration plane. --- README.md | 8 ++++++-- packages/platform/README.md | 4 ++-- 2 files changed, 8 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 360b73b..81146d5 100644 --- a/README.md +++ b/README.md @@ -8,9 +8,13 @@ **An open framework for agentic workflows. Bring your process, LLMs, scripts, agents, and infra — we handle the orchestration.** -Compose agent harnesses in YAML from LLM calls, scripts, validators, and transforms — wired with loops and conditionals. Mix deterministic and non-deterministic steps so your linter, tests, and schemas gate the LLM. Built-in support for Anthropic / OpenAI / Gemini / Ollama, coding-agent runtimes, SQLite / Postgres, OTel, and local / Docker / remote workers. Every layer has an extension point where the built-in doesn't fit. Scale agentic workflows like production infrastructure. +- **Compose** agent harnesses in YAML — LLM calls, scripts, validators, transforms, loops, and conditionals. +- **Gate the LLM with your tools** — deterministic steps (linters, tests, schemas) wrap non-deterministic LLM calls, so output is checked on every run. +- **Plug in your stack** — Anthropic, OpenAI, Gemini, Ollama, coding-agent runtimes. Every layer has an extension point where the built-in doesn't fit. +- **Run anywhere** — local, Docker, or remote workers; SQLite or Postgres; OTel-native. +- **Scale like infra** — multi-worker scheduling, approval gates, cost ceilings, live dashboard. -Ships with a reference SDLC template so you can see an end-to-end pipeline running in minutes. The framework is domain-agnostic: point it at code review, content generation, ops runbooks, data pipelines — anything where multiple LLM calls need to be coordinated with humans in the loop. +Ships with a reference SDLC template — runnable end-to-end in minutes. Domain-agnostic: point it at code review, content generation, ops runbooks, data pipelines — anywhere multiple LLM calls need to be coordinated with humans in the loop. > **For platform teams:** think of it as Kubernetes-style orchestration for agentic workloads — control plane, execution plane, declarative specs — without the cluster. diff --git a/packages/platform/README.md b/packages/platform/README.md index 071dfac..fc4ef20 100644 --- a/packages/platform/README.md +++ b/packages/platform/README.md @@ -1,8 +1,8 @@ # @mandarnilange/agentforge -**Production infrastructure for AgentForge.** Distributed execution, PostgreSQL, full observability, crash recovery. +**Production infrastructure for AgentForge.** Run agentic workflows at scale — distributed execution, PostgreSQL, full observability, crash recovery. -Extends [@mandarnilange/agentforge-core](https://www.npmjs.com/package/@mandarnilange/agentforge-core) with everything needed to run AI agent workflows at scale — Docker/remote executors, PostgreSQL persistence, OpenTelemetry tracing, rate limiting, and multi-node worker scheduling. +Extends [@mandarnilange/agentforge-core](https://www.npmjs.com/package/@mandarnilange/agentforge-core) — the open framework for agentic workflows — with everything needed to run it at scale: Docker / remote executors, PostgreSQL persistence, OpenTelemetry tracing, rate limiting, and multi-node worker scheduling. Bring your own LLMs, scripts, agents, and infra; this package handles the orchestration plane. > **Note:** This package requires `@mandarnilange/agentforge-core` as a peer dependency. The platform listing depends on it directly, so a single install pulls both: > ```bash From 5a1834a322193b7082952a4c87dadd7c5575e9d8 Mon Sep 17 00:00:00 2001 From: Mandar Nilange Date: Fri, 1 May 2026 22:37:49 +0530 Subject: [PATCH 2/2] docs: weave subtle K8s framing into platform README MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Add a Kubernetes analogy in two places where platform engineers will expect it and the analogy is genuinely load-bearing: 1. Lead paragraph (visible on npmjs.com) — closing phrase identifies the package as the "Kubernetes-style orchestration plane underneath" with control plane + capability-scheduled workers. 2. Architecture section — one-paragraph preamble explaining the control-plane / execution-plane separation, capability-based nodeAffinity scheduling, and the deployment topology. Audience is platform teams; the analogy teaches the design rather than serving as a marketing line. Hero one-liner stays clean. --- packages/platform/README.md | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/packages/platform/README.md b/packages/platform/README.md index fc4ef20..a8fb2b4 100644 --- a/packages/platform/README.md +++ b/packages/platform/README.md @@ -2,7 +2,7 @@ **Production infrastructure for AgentForge.** Run agentic workflows at scale — distributed execution, PostgreSQL, full observability, crash recovery. -Extends [@mandarnilange/agentforge-core](https://www.npmjs.com/package/@mandarnilange/agentforge-core) — the open framework for agentic workflows — with everything needed to run it at scale: Docker / remote executors, PostgreSQL persistence, OpenTelemetry tracing, rate limiting, and multi-node worker scheduling. Bring your own LLMs, scripts, agents, and infra; this package handles the orchestration plane. +Extends [@mandarnilange/agentforge-core](https://www.npmjs.com/package/@mandarnilange/agentforge-core) — the open framework for agentic workflows — with everything needed to run it at scale: Docker / remote executors, PostgreSQL persistence, OpenTelemetry tracing, rate limiting, and multi-node worker scheduling. Bring your own LLMs, scripts, agents, and infra; this package is the Kubernetes-style orchestration plane underneath — control plane + capability-scheduled workers, declarative specs, no cluster required. > **Note:** This package requires `@mandarnilange/agentforge-core` as a peer dependency. The platform listing depends on it directly, so a single install pulls both: > ```bash @@ -118,6 +118,8 @@ docker compose -f packages/platform/docker-compose.worker.yml up -d ## Architecture +Platform borrows Kubernetes' separation of concerns — a **control plane** decides what runs where, an **execution plane** of nodes runs it. Both ship in this package; deploy them together on one host or split across machines (see *Distributed* in Quick Start above). Nodes advertise capabilities (`llm-access`, `docker`, `high-memory`, `git`, …) and the scheduler matches each agent's `nodeAffinity` to the pool — same mental model as Pod scheduling. + Platform extends core via its ports/adapters pattern: ```